Advanced topics for group projects

Statistical Physics 653, Fall 2010, James P. Sethna

Last correction at September 7, 2018, 3:15 pm

Below find a brief discussion of applications of the Renormalization Group

to varied fields of physics and science. I’m also happy to entertain other
suggested topics.
(0)  expansion and 1/N expansion methods. Fluctuations become
less important in higher dimensions. Above an ‘upper critical dimension’
(usually four for non-disordered systems) continuous phase transitions can
be described by mean-field theory, ignoring the fluctuations. Wilson’s orig-
inal renormalization-group calculation was a perturbation about mean-field
theory, using dimension as a small parameter (the 4 −  expansion). Cardy
uses the “operator product expansion” (OPE) to implement this  expansion,
rather different from the Feynman-diagram methods used by most other re-
view articles and texts.
In 2009, Zach Lamberty and Darryl Ngai covered this topic. We could
also try to develop an exercise that implements the  expansion (or perhaps
a 1/N expansion) in a context where the algebra isn’t too messy and the
problem is accessible and interesting.
(1) Kosterlitz-Thouless and topological phases. Fluctuations be-
come more important in low dimensions. Indeed, below the ‘lower critical di-
mension’ for a given system the fluctuations drive the critical temperature to
zero. At the lower critical dimension, the phase transition often has unusual
scaling – not power laws, but exponentials in some quantities and sometimes
jumps in others. The freezing of crystals and superfluids in two dimensions,
along with fluctuations in the shapes of (three-dimensional) crystal surfaces,
are described by the Kosterlitz-Thouless transition. Much of the beautiful
physics of this transition (and especially the related 2D melting of Halperin
Nelson and Young) is hidden by the elastic theory and other details needed
before coarse-graining. Basically the same renormalization-group ideas can
be found with fewer complications in the one-dimensional Ising model with
inverse-square interactions.
In 2008 Ben Hunt presented theory and (Cornell) experiments on super-
fluid films, and Turan Birol and Wee Don Teo worked out and presented the
renormalization-group treatment of the inverse-square Ising model. (a) We
could try to develop an exercise that illustrates the features of topologi-
cal order – phase transitions without broken symmetries. (b) There are

1
deep connections between these classical systems and the quantum physics
of point defects coupled to Fermi gases: we could work out and present
Leggett’s theory of macroscopic quantum tunneling, and perhaps the Ander-
son/Yuval/Wilson theory of the Kondo effect, or other similar applications
(quantum tunneling of atoms off of STM tips?)
(2) Disordered and glassy systems. The study of dirt (random im-
purities and defects frozen into the system) and its effects both on phases
and phase transitions is a rich and controversial field of statistical mechan-
ics. Dirt can act to dramatically slow the dynamics (long-time tails, glassy
behavior), act to freeze the dynamics (spin-glass transitions), or lead to quali-
tatively new phenomena (localization of electron wave-functions). The effects
of randomness often can be separated into ‘typical’ (Imry-Ma) and ‘extreme’
(Griffiths) fluctuations.
In 2008, Robin Baur, YJ Chen, Vikram Gadagkar, and Joe Chen did
an in-depth treatment of the equilibrium behavior of the random-field Ising
model (RFIM), which illustrates dangerous irrelevant variables and leads,
by fairly simple arguments, to diverging barriers to relaxation. Vikram also
touched on the Harris criterion, which governs whether “random Tc ” disorder
is relevant or irrelevant at a phase transition. The RFIM is also a moral-
ity tale, with dueling experimental and theoretical views, abuses, politics,
and noble behavior, with a resolution that makes everyone correct and yet
both sides look sheepish. (a) We could move on to describe one of the many
other disordered systems – perhaps localization and the metal-insulator tran-
sition. (b) We could develop an exercise on diverging barriers, perhaps based
on the next-neighbor Ising model (“Prediction of Logarithmic Growth in a
Quenched Ising Model” Joel D. Shore, and James P. Sethna, Phys. Rev. B
43, 3782 (1991)).
(3) Hydrodynamics, fluctuations, flocking, and all that. New
physics often emerges on long length and time scales, even in the absence of
phase transitions. Hydrodynamics – originally the study of the flow of fluids
– has now been generalized to encompass the multiplicity of systems where
complex microscopic interactions lead to simple continuum laws. The general
approach is elegant: write down the most general theory allowed by symme-
try, expanding in powers of gradients and order parameters, and then keep
the most important terms. In the absence of noise terms, this procedure gen-
eralizes the familiar continuum limits we take in deriving the wave equation
and elastic theory. With noise, things get trickier, and renormalization-group
ideas come in.

2
 In 2008,, Gordon Berman discussed flocking – the motion of flocks of
animals. He noted in the end that the renormalization-group theory of the
“flocking transition” (where all the animals coordinate just enough to end up
migrating in the same direction) seems still confusing – perhaps abrupt and
not continuous. The flocking phase in two dimensions, though, is remarkable:
the existence of long-range order in 2D flocking would seem to violate the
Mermin-Wagner theorem . . . In 2010, Jesse Silverberg, Matt Bierbaum, and
Adam Holmes did experiments and simulations for human flocking behavior
at hard rock concerts, studying the transition between mosh pits and circle
pits. (a) There are theories of fluctuating rods and theories of traffic jams
that we could present, or even develop into exercises. (b) I have an exercise
on the wave equation that currently ignores noise, and we should see if we
can show that noise is irrelevant there.
(4) Fermi liquid theory and the renormalization group. Why are
metals described so well in terms of non-interacting electrons filling a Fermi
sea? The traditional theory – Green’s functions and quasiparticles, Fermi
liquid theory, and diagrams – is powerful and useful, but doesn’t tie in well
with modern statistical mechanics and field theory. A wonderful new ap-
proach treats the Fermi sea as a renormalization-group fixed point, where
the coarse-graining removes the high-energy electrons and holes, incorpo-
rating their screening effects in renormalized interaction strengths between
the quasiparticles. This renormalization-group is unusual, in that there are
an infinite number of marginal directions (neither relevant nor irrelevant).
These marginal directions are quadratically unstable to attractive interac-
tions, which are responsible for the transformation of metals at low temper-
atures into superconductors, spin-density waves and charge-density waves.
The transition temperature is neatly explained using the renormalization
group flows.
In 2008, Brian Daniels, Duane Loh, Mark Transtrum, and Johannes Lis-
chner presented a brief summary of the traditional Green’s function / quasi-
particle approach to Fermi liquid theory, an intuitive introduction to the
renormalization-group approach, a discussion of how to use quadratically
unstable modes to understand the BCS formula for the transition tempera-
ture of superconductors, and a whirlwind introduction to the technicalities
of coarse-graining the interacting Fermi liquid. In 2010, Mihir Khadilkar,
Shivam Ghosh, Kartik Ayyer, and Kyungmin Lee discussed both these theo-
ries and quantum phase transitions. (a) Duncan Haldane some twenty years
ago gave some talks about how the Fermi surface was the ‘order parameter’,

3
and I think that the quasiparticles somehow were related (via bosonization?)
to Goldstone modes. I don’t believe I ever heard these talks except second-
hand, and I don’t think these ideas were ever published, but we could think
deeply and see if we can make any sense of them.
(5) Quantum phase transitions. Quantum mechanics is usually not
needed to describe the exponents and asymptotic behavior near finite-temper-
ature phase transitions (quantum fluctuations are irrelevant perturbations at
finite-temperature critical points). However, there are phase transitions at
zero temperature (like the metal/insulator, superconductor/insulator, and
quantum Hall plateau transitions). More important, a transition at zero
temperature can dominate a large region of the finite-temperature phase di-
agram. The behavior right on the critical line might be classical, but (oddly)
the quantum effects can dominate farther away. Much recent activity, both
in high-Tc systems and in other quantum magnets, has been interpreted in
terms of quantum fluctuations due to a zero-temperature fixed point.
In 2008, Yang Xie gave a broad introduction to the experiments on high-Tc
superconductivity and the evidence suggesting an underlying quantum crit-
ical point. Hitesh Changlani, JeeHye Lee, Stefan Natu, and Ben Heidenre-
ich, Josh Berger introduced the connection between d-dimensional quantum
statistical mechanics and d + 1-dimensional classical statistical mechanics
and explained the scaling behavior of finite-temperature quantum critical
phenomena using universal crossover functions. In 2010, Mihir Khadilkar,
Shivam Ghosh, Kartik Ayyer, and Kyungmin Lee discussed both these the-
ories and Fermi liquid theory. (a) There is a somewhat separate discussion
of critical phenomena in quantum systems coupled to heat baths which goes
under the name of macroscopic quantum tunneling. How these ideas are
related are unclear, but might be worth pursuing.
(6) Conformal field theory. At a critical point, the effects of the lat-
tice disappear. An Ising model on a square lattice becomes both translation
and rotation invariant on long length scales. The symmetry of the fixed
point is larger than the symmetry of the original Hamiltonian: there is an
emergent rotational symmetry. In addition, another new symmetry arises,
incorporating changes of scale – a system at its critical point appears self-
similar. One can argue that combining these symmetries, for systems with
short-range interactions, naturally leads to a larger invariance under confor-
mal transformations – transformations that are rotations and dialations that
vary in space. In two dimensions, conformal invariance is a huge symmetry
group: basically one transformation for every analytic function on the com-

4
plex plane. In the study of string theory, it was discovered that the possible
critical points in two dimensions can be categorized by using this huge sym-
metry group. This is fundamentally our best explanation of why so many
two-dimensional critical points have exact solutions (with simple rational
values for critical exponents), while essentially no higher-dimensional models
are exactly solvable or have simple exponents (except when mean-field theory
applies).
Stochastic Loewner evolution. As a related technique, it has recently
been discovered by mathematicians that the interfaces at two-dimensional
critical points (between up and down in the Ising model, around the perco-
lation cluster, etc.) are all generated by a generalization of a random walk,
called stochastic Loewner evolution (SLE). Different models are described by
different values of a constant κ. This appears to be both cool and mathe-
matically powerful.
In 2008, Justin Vines, Phil Kidd, Ben Machta and Srivatsen Chakram
gave a cool presentation of conformal invariance and SLE, with software
distorting square simulations of the Ising model into weird shapes, software
generating the fractal boundaries for different κ, discussions of how to classify
all 2D theories and calculate exponents, and discussions of how these methods
can be used to calculate correlation functions. (a) Writing up these ideas into
a chapter would seem a great idea. (b) Transforming their software into a
general-purpose conformal transformation tool would be great. I envision
an exercise in every stat mech class where you distort an Ising model /
percolation model / etc. by any analytic function you care to type in, and
see that at Tc that you can’t tell where it’s been stretched. I have Justin’s
notes and PowerPoint, and a video DVD of one of the presentations.
(7) Recent developments in the theory of glasses. Glasses are dis-
ordered like liquids, but rigid like crystals. Unlike spin glasses and other
well-studied disordered systems, the disorder in glasses freezes in as the glass
becomes rigid. Fundamentally, there is no consensus about why glasses be-
come so viscous over such a small range of temperatures. While I still believe
it is possible and likely that glasses represent some kind of avoided thermo-
dynamic phase transition, current interesting theories describe it as a purely
dynamical phenomena. Jamming theories due to Andrea Liu and collabo-
rators (former Cornell grad), new analysis of kinetically constrained models
by Daniel Fisher and collaborators (Cornell child and undergrad), mode-
coupling theories by Jean-Philippe Bouchaud and collaborators, and now
some wonderful new curved-space simulations of Gilles Tarjus and collabo-

5
rators make this an exciting new area of research.
(8) Networks. There has been a flurry of high-profile papers analyzing
the structures of networks in the real world, and trying to duplicate the sta-
tistical structures and qualitative features using statistical mechanical mod-
els. Small world networks (“six degrees of separation”), random networks,
scale-free networks (with power-law distributions for the number of neighbor
links for a node), and grown networks are all topics that have scaling and
renormalization-group implications, as well as implications for the spread of
disease and the vulnerability of the Internet.
In 2009, Igor Segota, Tom Payne, and Alex Alemi discussed phase transi-
tions on networks, and disease propagation (with an entertaining simulation
of zombie dynamics).
(9) Chaos and turbulence. The renormalization group has been cen-
tral to our understanding of the onset of chaos. The period doubling route to
chaos, the quasiperiodic route to chaos that I studied as a post-doc, and the
breakdown of the last KAM torus that governs particle accelerators would
make a great topic for a class presentation. There are also some exciting
new statistical mechanics developments in the scaling behavior of fully de-
veloped turbulence (I remember something about conformal invariance or
string theory offshoots)?
(10) Depinning transitions. Earthquakes, magnetic Barkhausen noise,
plastic deformation of crystals, superconductor breakdown, charge-density
wave conduction, raindrops running down windshields – all are depinning
transitions, with an external force fighting against disorder, leading to a
jerky evolution. A rather unified picture has emerged, that would lend itself
nicely to a coherent presentation.
(11) Surface growth. The deposition of matter onto crystalline surfaces
(crystal growth from the melt, molecular beam epitaxy, diffusion limited ag-
gregation (DLA)) and the inverse problem of surface etching have a com-
plex, nonequilibrium dynamics that involves many length and time scales.
Faceting, deposition noise, ‘uphill’ currents from Ehrlich-Schwoebel barriers,
and shadowing effects lead to a rich theory. Scaling theories of island growth
and aggregation, KPZ and other statistical models, and the subtle multifrac-
tal behavior of DLA – all are fascinating problems where renormalization
group and scaling are involved.
(12) Spin glasses, neuroscience, and NP complete problems. Spin
glasses at low temperatures have no long-range order in space, but do have
long-range order in time: the spins are frozen into a disordered configura-

6
tion. Here there are two distinct theoretical threads. Mean-field theories
of spin glasses (with infinite-range interactions, representing the behavior
in spaces of large dimension) have a wonderfully rich structure. There is a
huge hierarchy of ground states (with an ultrametric topology), a diverging
nonlinear susceptibility, and exotic calculational tools (replica theory and
the cavity method) to solve for properties. Cluster models of spin glasses,
directly applicable to two and three-dimensional spin glasses, propose or
posit only two distinct ground states but study a variety of metastable states
formed by flipping clusters of spins with competing interactions. While one
could argue that the cluster models may be closer to real spin glasses, the
mean-field models have found surprising applications, first in neuroscience
(where each neuron is connected to a huge number of neighbors) and now
in computational complexity. Indeed, the most successful algorithm known
for solving one class of NP complete problems (the most challenging class of
computational questions – including the traveling salesman and map color-
ing problems) is based on the cavity method designed to study spin glasses.
Little of this directly involves the renormalization group, but scaling and
critical phenomena ideas pervade the subject.
In 2009, Elijah Bogart, Poornima Padmanabhan and Hitesh Changlani
discussed NP-complete problems in computer science, and their relations to
the random energy model and the number partitioning problem.
(13) Wetting. A drop of liquid on a solid surface forms a bead – the
liquid, solid, and vapor coexistence at the line of intersection. The contact
angles depend on the three surface tensions (liquid-solid, liquid-vapor, and
solid-vapor), minimizing the total free energy. If the liquid and solid like one
another, the liquid drop will become thin and the contact angle large. At
complete wetting, the minimum free energy spreads the drop out as a mono-
layer on the surface. This often happens if the liquid and vapor (or two liquid
phases) are near a critical point where they become indistinguishable (and
where the interfacial energy cost for the liquid-vapor cap of the droplet goes
to zero). Complete wetting is thus often a precursor surface phase transition;
Ben Widom and Carl Franck here studied it in great detail theoretically and
experimentally earlier in their careers.
(14) Equilibrium crystal shapes. Most faceted crystals (like dia-
monds) are not in equilibrium – the crystals cleave along low energy sur-
faces, but the atoms have not had time to rearrange between different facets
to minimize their energy. The study of crystals that have reached this shape
evolution, though, is fascinating. The existence of facets demands a rough-

7
ening transition (in the same universality class as the Kosterlitz-Thouless
transition discussed in (1)). The evolution and coarsening of faceted surfaces
has a nice scaling theory. The construction of the equilibrium shape from the
surface free energy (the Wulff construction) is a multidimensional Legendre
transformation. Continuous facet edges have scaling properties determined
by two-dimensional (exactly solvable) models. Combining these topics with
a renormalization-group theme could be a pretty subject.
In 2009, Kendra Weaver covered the equilibrium crystal shape theory,
introducing the Wulff construction. The connection with Kosterlitz-Thouless
and roughening, and the connection between facet edges and critical points,
remain open problems.
(15) Asymptotic analysis of partial differential equations. Asymp-
totic analysis is an art, not a science. The text of Bender and Orszag pro-
vides an introduction to the many sneaky methods one can use to find the
asymptotic behavior of various mathematical quantities as parameters get
large or small. Nigel Goldenfeld introduced some wonderful new methods
for solutions of nonlinear partial differential equations, drawing on ideas and
language used in the renormalization-group analysis of phase transitions. He
provides an introduction in Chapter 10 of his text, Lectures on Phase Transi-
tions and the Renormalization Group, Frontiers in Physics, Addison-Wesley,
1992.
(16) Bifurcation theory. Much of the theory of dynamical systems
(the time evolution of ordinary differential equations) focuses on bifurcation
theory, the study of qualitative changes in the dynamics as system parame-
ters evolve. Continuous bifurcations are amazingly analogous to continuous
phase transitions, except that they describe systems with only a few active
degrees of freedom. Dynamical systems theory lumps together bifurcations
into natural families: saddle-node bifurcations, pitchfork bifurcations, Hopf
bifurcations, . . . These families each have their normal form – analogous to
universality classes. A famous theorem shows that any Hopf bifurcation, say,
can be transformed into the normal form Hopf bifurcation by a continuous
change of coordinates. This theorem has its analogy in the analytic correc-
tions to scaling needed to understand scaling phenomena at critical points.
In 2010, Alex Moore and Maicol Ochoa Daza did a presentation on these
connections, but did not get into analytic corrections to scaling.
(17) Droplet breakup. Dripping faucets have hidden universal scaling
behavior. As the drops pinch off, the breakup can be wonderfully compli-
cated, with the asymptotic form of the fluid interface forming shapes that

8
are independent not of the microscopic parameters but (in a reversal of the
usual scenario) of the macroscopic way the droplet was formed and exuded.
Itai Cohen has been working on this topic since his graduate student years...